Monte Carlo Path Tracer: CUDA vs CPU Implementation

This repository contains two implementations of a path tracer: one using CUDA for GPU acceleration and another using CPU multi-threading. The comparison demonstrates the significant performance advantage of GPU-based path tracing for real-time rendering.

CUDA Path Tracer (160 FPS)

CPU Path Tracer (10~25 FPS)

Project Structure

.
├── cuda_pathtracer/    # GPU-accelerated implementation
└── cpu_pathtracer/     # CPU-based implementation

Scene Preview

[Insert screenshot of your rendered scene here]

The scene features:

• A Cornell box with colored walls (red and green sides)
• Multiple spheres with different materials:
  • Glass (refractive)
  • Metal (reflective)
  • Glossy (mixed specular/diffuse)
• Multiple colored lights for interesting illumination effects
• Real-time camera movement and interaction

Performance Comparison

CUDA Path Tracer

• Frame Rate: 160 FPS on 4060 GPU
• Resolution: 1280x720
• Features:
  • Real-time rendering
  • Interactive camera movement
  • Progressive refinement
  • Multiple material types (Diffuse, Glossy, Specular, Glass)
  • Colored lighting effects
  • Monte Carlo path tracing with importance sampling

CPU Path Tracer

• Frame Rate: 10~20 FPS
• Resolution: 1280x720
• Features:
  • Same features as CUDA version but significantly slower
  • Multi-threaded rendering using OpenMP
  • Progressive refinement (but much slower accumulation)
  • Demonstrates why GPU acceleration is necessary for real-time path tracing

Why CUDA is Necessary

The CPU version demonstrates why GPU acceleration is essential for real-time path tracing:

1. Parallelization: Path tracing is inherently parallel - each pixel can be computed independently
2. Ray-Object Intersection: GPU can perform thousands of intersection tests simultaneously
3. Monte Carlo Sampling: Random sampling and light calculations benefit from massive parallelization
4. Memory Access: GPU's high memory bandwidth is crucial for accessing scene data

Building and Running

CUDA Version

mkdir build && cd build
cmake ..
make
./cuda_pathtracer/CUDAPathTracer
./cpu_pathtracer/CPUPathTracer

Controls(only CPU version)

• W/A/S/D: Move camera
• Mouse: Look around
• Tab: Toggle mouse capture
• Escape: Exit

Requirements

CUDA Version

• CUDA Toolkit 11.0 or higher
• NVIDIA GPU with Compute Capability 6.0 or higher
• C++17 compatible compiler
• CMake 3.18 or higher
• X11 development libraries(Linux)

CPU Version

• C++20 compatible compiler
• CMake 3.18 or higher
• OpenMP
• X11 development libraries

Technical Details

Both implementations use:

• Monte Carlo path tracing
• Multiple importance sampling
• Real-time progressive refinement
• Various material types (diffuse, glossy, specular, glass)
• Scene management with implicit objects
• Interactive camera controls(Cuda version has bugs)